one question about knowledge chess engine match one ply ?

[Karmazen & Oliver]

one question about knowledge chess engine match one ply ?

I was thinking of making a thematic party on knowledge, using fixed depth of a ply, another with 3 plys and others with 5 plys. perhaps another with 7 plys. no more that 9 or 11.

I ask:

using a stable depth would Be annulled the advantages of code optimization, code parts that anything has to do with the knowledge, but alone with the gross force and certain dose of luck when selecting many branches...

if we use those low levels of I calculate of plys, we would not find the engines that have a knowledge bigger than the positions and they choose the road but or less correct, or rather, they choose a better road that their rival, although this it is incorrect, depending on the given position.

I will prove it. go to 1 ply. if this is error conceptual i try 3, 5, 7, 9. no more.

no books. too.

[Karmazen & Oliver]

the match is running now, on arena. chessbase not permit do a tournment with plys stable. only times.

I see some interesting thins.

arasan lost all games for now.

r 3 is win a lot of.

r1beta take up MATE #to hiarcs 11 ¡

and zappa lost one piece on open game on two games.

I am seeing that the selection that makes the engine in the first place decants its later destination.

for that EVIDENTLY, in ply 1 choose a move-play, if it is bad, he/she stays with its, then in ply 2 chosse a move-play, if it is bad, it returns behind and he/she chooses another.

if the first selection was incorrect, the engine loses time. for that reason rybka arrives more far in the analyses.

tiger don´t permit me ply selection.

fruit play good selection too...

I hope see some conclusions. but I think that exits some interesting reflexion in this experiment...

I see some bugs on selection on 1 ply ? these bugs...

these errors can be dragged to the whole analysis chain... being based on first imprecise, or chancy selections, leaving to the engine to a road without exit.

many changes of pieces and an aggressive game are also observed.

[Karmazen & Oliver]

Do you know each engine reports ply differently? For example a Rybka 3 ply could be an Arasan 6 ply. A better experiment Uri Blass has tested playing his engine 10 ply against Rybka 3 ply ( I am not sure of the numbers he has posted this before you could check it), so you know how much handicap an engine could give to another and develop a testing procedure.

that example is incorrect.

if you permit that engine thinking more that one ply, you permit using optimization code, on selection 2 and 3 plys.

I see that R3 is win games versus... fruit, hiars 11, frtiz 9, and now r3 human play versus r3 dinamic... cpu, etc..

I see that it´s one ply, because engines play 12 games and all are indentical... only is necesary two games, one with black and other with white.

I think that exits some idea interesting about r3 have a lot of ... diferent.

[Uri Blass]

You are going to find nothing about knowledge by fixed depth because different engines search different lines at fixed depth.

Rybka clearly see more tactics at what she calls small depth so fixed depth shows nothing about strength of the evaluation of the engine.

Here is an example

I believe that Rybka finds b4 with a positive score thanks to search and not thanks to evaluation

Rybka 3 1-cpu 32-bit - so k
[D]4k3/pppp1p2/8/b4p2/P3P3/2PP4/1P3PPP/4K3 w - - 0 1

Analysis by Rybka 3 1-cpu 32-bit :

1.b2-b4
± (0.86) Depth: 1 00:00:00
1.b2-b4 f5xe4
± (0.86) Depth: 1 00:00:00

(so k, 05.09.2008)


Uri

[Karmazen & Oliver]

You are going to find nothing about knowledge by fixed depth because different engines search different lines at fixed depth.

Rybka clearly see more tactics at what she calls small depth so fixed depth shows nothing about strength of the evaluation of the engine.

Here is an example

I believe that Rybka finds b4 with a positive score thanks to search and not thanks to evaluation

Rybka 3 1-cpu 32-bit - so k
[D]4k3/pppp1p2/8/b4p2/P3P3/2PP4/1P3PPP/4K3 w - - 0 1

Analysis by Rybka 3 1-cpu 32-bit :

1.b2-b4
± (0.86) Depth: 1 00:00:00
1.b2-b4 f5xe4
± (0.86) Depth: 1 00:00:00

(so k, 05.09.2008)


Uri

I think that you are very confuse... sorry me but:

do apologize, but for that you think that that play doesn't bear evaluation?

if I was programmer, one of the first watchwords to evaluate serious good movements that besides maximizing the I number of stalls that attack my pieces, my plays they will be good if they minimize the reach of the rival's pieces, evidently b4 ¡ for that besides winning space (knowledge?), do you attack a contrary piece, (do think like a single play), does the engine think: " if it doesn't move it, I eat up it... ; -) and if it moves it... their bishop threatens less stalls and peon increases his dynamic value...

in the following play, the thought repeats, rybka will have programmed that it should attack to the pieces rivals so that they don't have available plays, (knowledge lives), using the proverb: "if it is bad for my enemy, it will be good for my"...

all that is knowledge.

also... continue reading the following one post. , although rybka wins in almost all the parties, depending the knowledge of the engine, it takes him more time in moves-plays, to be able to win it.

ok? see:

http://img143.imageshack.us/my.php?imag ... ctiur1.jpg


The programers only need see how rybka select his moves on 1 plys and do the algoritms similar efect.. and after this test on "test positions, nuhm, marathon, problems. etc...( see in diferents positions )"

it´s easy don´t lost all games versus r, win maybe will be dificult. but LOST ? always ¡ never ¡¡¡

I think that the first selection move is BAD, all moves continue can be bad, because the algorit is recursive and always use the same concepts...

BUG... all word have BUGS ¡ enormous initials bugs.

(sorry for programers... , but I think that something is incorrect on 1 plys in a lot of engines.)

bye, from spain. OLIVER:

[Karmazen & Oliver]

interesting:

tomorrow show results. rh versus rc and rd.

R dinamic wins on 17 moves ¡ to r cpu.

rybka dinamic play fast win with white colors.

rybka human is more slowly to win with white, but is more sure when he/she is playing with black.

The best engine is Rybka human ? NO books, after I would do a test num match with alternetive colors to see which is better...

http://img225.imageshack.us/my.php?imag ... yonqt6.jpg


this is very surprising, at first sight one wait that with A single play of depth, no engine loses the game, surprisingly all lose against rybka.

in reference to the chess knowledge, it is curious to point out that depending on the fritz version, f6, f7, f8, f9, rybka needs more plays-moves to finish winning. of all the fritz, the more difficult for rybka it has been deep fritz 7 that I need 50-60 moves to win it...

tiger take to goods results, but is possible that this engines play with to limit 3 plys. ( by program options limits) I am not sure... ch14 versus ch15 : +1, 0 = -1, with the same moves. 40 for win ch14 and 40 moves too to win ch14, when they play among them.

I believe that all the engines has serious trial problems, from the first play-moves selecion. ???

http://img143.imageshack.us/my.php?imag ... ctiur1.jpg


OK. 1 ply don´t permit use optimization code. only selection chess knowledge, why some engines lost one piece in only one move ¡? this is very surprise.

NO TIME limit. don´t problems. OK. 1 ply don´t permit use optimization code. only selection chess knowledge, why some engines lost one piece in only one move ¡? this is very surprise.

NO TIME limit. don´t problems. why do some engines makes those simple trial errors?

The programers only need see how rybka select his moves on 1 plys and do the algoritms similar efect.. and after this test on "test positions, nuhm, marathon, problems. etc...( see in diferents positions )"

it´s easy don´t lost all games versus r, win maybe will be dificult. but LOST ? always ¡ never ¡¡¡

I think that the first selection move is BAD, all moves continue can be bad, because the algorit is recursive and always use the same concepts...

BUG... all word have BUGS ¡ enormous initials bugs.

(sorry for programers... , but I think that something is incorrect on 1 plys in a lot of engines.)

bye, from spain. OLIVER:

[Uri Blass]

You are going to find nothing about knowledge by fixed depth because different engines search different lines at fixed depth.

Rybka clearly see more tactics at what she calls small depth so fixed depth shows nothing about strength of the evaluation of the engine.

Here is an example

I believe that Rybka finds b4 with a positive score thanks to search and not thanks to evaluation

Rybka 3 1-cpu 32-bit - so k
[D]4k3/pppp1p2/8/b4p2/P3P3/2PP4/1P3PPP/4K3 w - - 0 1

Analysis by Rybka 3 1-cpu 32-bit :

1.b2-b4
± (0.86) Depth: 1 00:00:00
1.b2-b4 f5xe4
± (0.86) Depth: 1 00:00:00

(so k, 05.09.2008)


Uri

I think that you are very confuse... sorry me but:

do apologize, but for that you think that that play doesn't bear evaluation?

if I was programmer, one of the first watchwords to evaluate serious good movements that besides maximizing the I number of stalls that attack my pieces, my plays they will be good if they minimize the reach of the rival's pieces, evidently b4 for that besides winning space (knowledge?), do you attack a contrary piece, (do think like a single play), does the engine think: " if it doesn't move it, I eat up it... ; -) and if it moves it... their bishop threatens less stalls and peon increases his dynamic value...

in the following play, the thought repeats, rybka will have programmed that it should attack to the pieces rivals so that they don't have available plays, (knowledge lives), using the proverb: "if it is bad for my enemy, it will be good for my"...

all that is knowledge.

also... continue reading the following one post. for that sorpredentemente, although rybka wins in almost all the parties, depending the knowledge of the engine, it takes him more time in plays, to be able to win it.

ok? see:

http://img143.imageshack.us/my.php?imag ... ctiur1.jpg

The point is not only the move
Rybka already see that it wins the bishop at depth 1
other programs do not see it.

It cannot be result of evaluation and is clearly result of search.

Rybka search more lines when you tell it to search depth 1.
You cannot tell rybka to search depth 0 or depth -1 or depth -2 under Fritz gui but it is possible to do it under console mode


uci
id name Rybka 3 32-bit
id author Vasik Rajlich, Larry Kaufman
option name UCI_Query type button
option name Hash type spin min 2 max 4096 default 32
option name Max CPUs type spin min 1 max 2048 default 2048
option name NalimovPath type string default <empty>
option name NalimovCache type spin min 1 max 256 default 1
option name NalimovUsage type combo default Rarely var Frequently var Normally v
ar Rarely var Never
option name Preserve Analysis type check default false
option name Clear Hash type button
option name Saved Hash File type string default <empty>
option name Save Hash type button
option name Load Hash type button
option name Persistent Hash Enabled type check default false
option name Persistent Hash File type string default <empty>
option name Persistent Hash Play Depth type spin default 64 min 8 max 64
option name Persistent Hash Write Depth type spin default 10 min 8 max 64
option name Persistent Hash Size type spin default 16 min 8 max 65536
option name Persistent Hash Reset type button
option name Persistent Hash Resize type button
option name Persistent Hash Merge File type string default <empty>
option name Persistent Hash Do Merge type button
option name Ponder type check default false
option name MultiPV type spin default 1 min 1 max 100
option name MultiPV_cp type spin default 32768 min 0 max 32768
option name Contempt Play type spin default 15 min -250 max 250
option name Contempt Analyze type spin default 0 min -250 max 250
option name UCI_LimitStrength type check default false
option name UCI_Elo type spin default 1200 min 1200 max 2400
option name UCI_EngineAbout type string default www.rybkachess.com
option name Time Buffer type spin default 0 min 0 max 900
uciok
position fen 4k3/pppp1p2/8/b4p2/P3P3/2PP4/1P3PPP/4K3 w - - 0 1
go depth -2
info depth 1 score cp 52 time 1 nodes 2 nps 2048 pv e4f5
info depth -2 time 1 nodes 4 nps 4096
bestmove e4f5 ponder a5c3

[Karmazen & Oliver]

The point is not only the move
Rybka already see that it wins the bishop at depth 1
other programs do not see it.


It cannot be result of evaluation and is clearly result of search.

Rybka search more lines when you tell it to search depth 1.
You cannot tell rybka to search depth 0 or depth -1 or depth -2 under Fritz gui but it is possible to do it under console mode

incorrect ?

although was not 1 ply exactly, that that if we know it is that he/she obtains the result with very little depth and analyzing single 50-200 positions.

Although that is this way, he/she doesn't stop to be curious that using the same engine, r human, r dinamic, r cpu, (all will work to the same depth, be this 1 or 1,5 or 2, that is irrelevant...) ( only chesstiger maybe use 3 plys.. and win versus rybka ¡... and the last version on chesstiger 2007 don´t permit me using plys selection level...)

it is also curious that ALWAYS, they carry out the same plays, then the factor time not this implied in the selections and the searches... they don't always report the same solution, random variables exist when a search depth is applied. (good lucky... and this permit that engines don´t play always similar moves)

the fact that they always play the same plays and that the victories always obtain them the engines more outposts mean that they are basing on knowledge you position them and strategic.

Code: Select all

USING ARENA chess GUI:

   Motor                     Puntaje   Ry  Ry  Ry  Ry  Fr  Ry  Za  Hi  Ar    S-B
1: Rybka 3 Human 1-cpu w32   21,5/24  ··· 101 =1= 1=1 111 111 111 111 111  212,75
2: Rybka 3 Dynamic 1-cpu w32 20,5/24  010 ··· 101 111 111 1=1 111 111 111  196,25
3: Rybka 3 1-cpu w32         18,0/24  =0= 010 ··· 101 101 111 111 111 111  158,50
4: Rybka v2.3.2a.w32         16,5/24  0=0 000 010 ··· 111 111 111 111 111  123,25
5: Fruit-2-3-1               11,0/24  000 000 010 000 ··· 010 111 111 111   58,50
6: Rybka v1.0 Beta.w32       10,5/24  000 0=0 000 000 101 ··· 101 111 111   56,75
7: Zappa                     5,5/24   000 000 000 000 000 010 ··· =1= 1=1   19,75
8: Hiarcs11MP                4,0/24   000 000 000 000 000 000 =0= ··· 111    7,00
9: Arasanx                   0,5/24   000 000 000 000 000 000 0=0 000 ···    2,75

108 partidas jugadas / Torneo finalizado
Nombre del torneo: Arena tournament
Lugar/ País: PC-820D-OLIV, España
Nivel: 1 Medios mov.
Hardware: Dual        Intel(R) Pentium(R) D CPU 2.80GHz  with 2.048 MB Memory
Sistema operativo: Microsoft Windows XP Professional Service Pack 2 (Build 2600)
Archivo PGN: C:\Ajedrez\arena199beta5\Arenat 2007.pgn
Website: 
Correo electrónico: 

observe the positions, too many chances, the order is almost perfect and using fritz gui, engine TO versus engine B on 24 games 22 games is duplicated. for that are limiting the depth, this way we SEE the exact selections of the program algoritm, depending the analysis position.


and that will be key in the later searches. The error this in the principle, there is where it wins rybka.

truth that yes?

[bob]

one question about knowledge chess engine match one ply ?

I was thinking of making a thematic party on knowledge, using fixed depth of a ply, another with 3 plys and others with 5 plys. perhaps another with 7 plys. no more that 9 or 11.

I ask:

using a stable depth would Be annulled the advantages of code optimization, code parts that anything has to do with the knowledge, but alone with the gross force and certain dose of luck when selecting many branches...

if we use those low levels of I calculate of plys, we would not find the engines that have a knowledge bigger than the positions and they choose the road but or less correct, or rather, they choose a better road that their rival, although this it is incorrect, depending on the given position.

I will prove it. go to 1 ply. if this is error conceptual i try 3, 5, 7, 9. no more.

no books. too.

It also won't prove a thing. Some programs get way more out of 9 plies than others. For example, most current programs will do 9 plies instantly, while a program like ChessMaster (the king) might take minutes (or sometimes hours) to get that deep. plies are not uniform. Some programs extend much more aggressively than other s and equal plies will see those taking a lot more time. Other programs reduce more aggressively and equal plies will see those programs take far less time. you end up with results that are meaningless...

[Uri Blass]

I can prove that rybka's good moves at depth 1 is thanks to search and not thanks to evaluation

Here is another example

[D]rn3k1r/pp6/8/3N4/8/RP6/P7/K7 w - - 0 1

Rybka at depth 1 plays b4
programs that evaluate tactical threats could see that the black rook is trapped after Nc7 and play this move at depth 1

Rybka usually finds Nc7 at depth 1 thanks to search but not here because many checks of black make the winning of the rook too deep for rybka to find.

If I add one pawns to both sides rybka can find Nc7 at depth 1

[D]rn3k1r/pp5p/8/3N4/8/RP6/P6P/K7 w - - 0 1

Uri